Add instructions for using custom yaml config for collective benchmark

Author: raymondzouu

microbenchmarks/trillium/collectives/README.md

@@ -36,4 +36,23 @@ Results will be printed out and also stored at `/tmp/microbenchmarks/collectives
 gsutil cp -r /tmp/microbenchmarks/collectives gs://<your-gcs-bucket>
 ```
 
+### Run with a custom yaml config
+If you would like to run with a custom defined yaml with modified configurations (e.g. warmup_tries, tries, matrix_dim_range) you may do so by uploading it to a GCS bucket, pulling the yaml file from GCS in the workload, and then referencing the yaml file in the benchmark command. 
 
+Start by creating a yaml file `your_config.yaml`. Take a look at [1x_v6e_256.yaml](https://github.com/AI-Hypercomputer/accelerator-microbenchmarks/blob/35c10a42e8cfab7593157327dd3ad3150e4c001d/configs/1x_v6e_256.yaml) for an example yaml config. Then upload it to your GCS bucket:
+```
+gsutil cp your_config.yaml gs://<your-gcs-bucket>
+```
+
+Then use a modified launch command that pulls the yaml file from GCS and references it in the benchmark command:
+```
+python3 ~/xpk/xpk.py workload create \
+    --cluster=${CLUSTER_NAME} \
+    --project=${PROJECT} \
+    --zone=${ZONE} \
+    --device-type=v6e-256 \
+    --command="git clone https://github.com/AI-Hypercomputer/accelerator-microbenchmarks.git && cd accelerator-microbenchmarks && git checkout trillium-collectives && pip install -r requirements.txt && echo '4096 41943040 314572800' > /proc/sys/net/ipv4/tcp_rmem && export LIBTPU_INIT_ARGS='--megascale_grpc_premap_memory_bytes=17179869184 --xla_tpu_enable_sunk_dcn_allreduce_done_with_host_reduction=true' && gsutil cp gs://<your-gcs-bucket>/your_config.yaml configs/ && python src/run_benchmark.py --config=configs/your_config.yaml" \
+    --num-slices=1 \
+    --docker-image=us-docker.pkg.dev/cloud-tpu-images/jax-stable-stack/tpu:jax0.5.2-rev1 \
+    --workload=${WORKLOAD_NAME}
+```